Closed JaGarRod closed 3 years ago
Thanks you're right this is a discrepancy. I'd greatly welcome your contributions, and it's quite easy to do actually because you we have this in surv.glmboost
so you can just copy/paste!
https://github.com/mlr-org/mlr3extralearners/blob/main/R/learner_mboost_surv_glmboost.R
No requirement to do so but would be great if you could do the same for all of:
I will first try to finish the pull request for regr_glmboost and once that works I will complete the rest when I find some time
To avoid mistakes like mine in /issues/79, I see the following options:
I would suggest 2. or 3. @RaphaelS1 , let me know what you think is best and I will update the learner. Afterwards I will continue with the other three.
I don't think 4 is feasible but 3 sounds good. Arguably we don't even need a warning because as you demonstrated, one would naturally expect the family to be custom if a custom one is supplied. Would you mind also updating surv as well?
I'd also recommend adding an example to each to demonstrate how to use a custom family
See #82
Currently regr.glmboost learner has a parameter family which is of factor type and has to be chosen among 7 predetermined families. Can we introduce support to add a custom family as a parameter?
I would be happy to contribute if some guidance is provided.
To make it backwards compatible, should we create a customFamily parameter that depends on family parameter being FALSE or empty or "custom"?